neptune-scale

What it is and what it does

Neptune-scale is a client library for the Neptune experiment-tracking platform, designed to capture and send training metadata—metrics, configurations, files, and histograms—from your training loop to a centralized web dashboard. It sits between your training code and the Neptune backend (via neptune-api), handling the collection, buffering, and transmission of experiment data.

You initialize a Run object, call logging methods like log_metrics() and log_configs() during training, and optionally upload files or histograms. The library depends on standard utilities like requests, aiofiles, GitPython, and click to manage I/O, retries, and CLI interactions. It's built for foundation model training workflows where you need to monitor many per-layer signals without lag.

Use it for:

• Log training metrics (loss, accuracy) and hyperparameters from a model training loop to track experiment progress.
• Upload dataset samples, model checkpoints, or debug logs as files to Neptune for post-training inspection.
• Monitor per-layer activations, gradients, and weight histograms during deep learning training to diagnose training instability.
• Tag and organize multiple training runs for comparison and grouping within a shared Neptune workspace.
• Debug training issues by drilling into logged metrics and files without re-running experiments.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Neptune-scale is a Python client library for logging and monitoring experiment metadata—metrics, configurations, files, and histograms—during model training, with integration to the Neptune web platform for visualization and analysis.

Yes, if you are training foundation models or large neural networks and want centralized experiment tracking with minimal setup. The low install friction, permissive license, and no known vulnerabilities make it safe to adopt. However, the aging maintenance status (261 days since last release) suggests you should verify that updates and support align with your project timeline before committing to it as a long-term dependency.

Install

neptune-scale on PyPI

pip install neptune-scale

uv add neptune-scale

poetry add neptune-scale

Installing neptune-scale

Before you install

Low install friction with a pure-Python wheel and 11 runtime dependencies that are all standard data/networking libraries. Maintenance status is aging—last commit 2026-01-19, 261 days since the latest release—so updates and bug fixes may lag.

License in practice

Apache-2.0 is permissive and poses no restrictions on use or redistribution; you can incorporate this library into commercial or proprietary projects without license obligations.

Quickstart

pip install neptune-scale

from neptune_scale import Run

run = Run(experiment_name="MyExperiment")
run.log_configs({"learning_rate": 0.001})
run.log_metrics(data={"loss": 0.17}, step=0)
run.close()

Requires NEPTUNE_API_TOKEN environment variable set; optionally NEPTUNE_PROJECT for project path. Python 3.9 or later.

Verify before relying

• Whether the aging maintenance status (261 days since release) affects stability or feature parity with the main Neptune product.
• Performance characteristics when logging thousands of per-layer metrics at scale, as claimed in the description.

Package facts

Evidence: neptune_scale-0.30.0-py3-none-any.whl

Keywords: MLOps, ML Experiment Tracking, ML Model Registry, ML Model Store, ML Metadata Store

Tags

More Python Modules packages